Improved multi model predictive control for distillation column

Abstract

Model predictive control (MPC) strategy is known to provide effective control of chemical processes including distillation. As illustration, when the control scheme was applied to three linear distillation columns, i.e., Wood-Berry (2x2), Ogunnaike-Lemaire-Morari-Ray (3x3) and Alatiqi (4x4), the results obtained proved the superiority of linear MPC over the conventional PI controller. This is however, not the case when nonlinear process dynamics are involved, and better controllers are needed. As an attempt to address this issue, a new multi model predictive control (MMPC) framework known as Representative Model Predictive Control (RMPC) is proposed. The control scheme selects the most suitable local linear model to be implemented in control computations. Simulation studies were conducted on a nonlinear distillation column commonly known as Column A using MATLAB® and SIMULINK® software. The controllers were compared in terms of their ability in tracking set points and rejecting disturbances. Using three local models, RMPC was proven to be more efficient in servo control. It was however, not able to cope with disturbance rejection requirement. This limitation was overcome by introducing two controller output configurations: Maximizing MMPC and PI controller output (called hybrid controller, HC), and a MMPC and PI controller output switching (called MMPCPIS). When compared to the PI controller, HC provided better control performances for disturbance changes of 1% and 20% with an average improvement of 12% and 20% of the integral square error (ISE), respectively. It was however, not able to handle large disturbance of + 50% in feed composition. This limitation was overcome by MMPCPIS, which provided improvements by 17% and 20% of the ISE for all of types and magnitudes of disturbance change. The application of MMPCPIS on a single model MPC strategy produced almost similar performance for both types of disturbances, while its application on MMPC yielded better results. Based on the results obtained, it can be concluded that the proposed HC and MMPCPIS deserve further detailed investigations to serve as linear control approaches for solving complex nonlinear control problems commonly found in chemical industry